Acoustic characteristics of disguised speech: speaker strategies and listener error patterns

A group of 13 participants were recorded in two conditions: 1) speaking normally and 2) altering speech to conceal their identity (i.e., disguised speech). Participants were not instructed how to disguise their speech because we were interested in which changes they would choose. A group of inexperienced listeners were largely inaccurate in matching participants' disguised speech to their normal speech. The largest changes between normal and disguised speech were in speaking rate, the first formant, fundamental frequency, and intensity. When listeners made correct matches, the pairs were similar in speaking rate and intensity, as shown by significant correlations. Incorrectly matched pairs were not significantly correlated, suggesting that listeners were not making good use of acoustic cues during those decisions. Overall, the third formant (F3) and speaking rate appeared to be useful acoustic indicators of identity when matching normal and disguised speech samples. Of those two variables, F3 was apparently underutilised by listeners. The implications for what spontaneous speakers do to disguise their speech and what naive listeners attend to when identifying disguised voice are discussed.